Volumetric datasets are found in many fields, such as engineering, material sciences, medical imaging, astrophysics. The exploration of volumetric datasets is not trivial, and is heavily impacted by the specific needs of users. In most airports for example, security agents deal with such data exploration in the context of baggage inspections. X-ray and tomography are two commonly used fluoroscopic scanning systems. X-ray systems provide a flattened 2D luggage scan while tomography systems produce transversal scans, also called slices. Thanks to data processing techniques such as the Radon transform, these systems can produce a full 3D scan, comprising a set of voxels with corresponding density data. Since the resulting X-ray scanned image only contains voxel or pixel densities, it cannot display the original material colours. The standard colour visual mapping uses three different colours (orange, green, and blue) to display the data density. Orange colour corresponds to low density (mainly organic items). In opposition, blue colour is used for high density values (i.e. metal). In the case of X-ray systems, green colour corresponds to the superposition of different kinds of materials or average density materials.
FIG. 1 demonstrates some of the ways in which an article may be obscured in a scan. As shown in FIG. 1, the displayed 2D scanned image can suffer from four issues:
Superposition: A threat (e.g. prohibited object like knife, cutter . . . ) may be sheltered behind dense materials. Sometimes, it's possible to see through this blind shield using functionalities such as high penetration (enhanced X-ray power) or image processing (contrast improvement). As shown in FIG. 1, the umbrella and dense collection of objects in the upper right hand corner 101 may obscure articles of interest.Location: Depending on its location inside the luggage, a threat can be difficult to detect. Objects located in the corners, in the edges or inside the luggage's frame are very difficult to identify. As shown in FIG. 1, the retractable trolley bars and the rigid corners of the case 102 may obscure articles of interest.Dissociation: Another way to dissimulate a threat is to separate and to spread parts of it in the luggage (weapons or explosives are composed of many separated items like the trigger, the barrel . . . ). This dissociation can be combined with other dissimulation techniques. As shown in FIG. 1, a number of apparently non-descript items 103 are present which are unlikely to attract particular attention, but which may be assembled to form some article of interest.Lure: An ill-intentioned individual may use a lure to hide the real threat. For instance, a minor threat like a small scissors may be clearly visible and catch security agent's attention while a more important threat remains hidden. As shown in FIG. 1, the metal rod 104 may attract the attention of the user, drawing it away from some less visible threat.
Volumetric data exploration with direct volume rendering techniques is of great help to visually extract relevant structures in many fields of science: medical imaging, astrophysics and more recently in luggage security. To leverage this knowledge extraction, many techniques have been developed. A number of existing basic technologies are known in this field, including volume visualization, transfer function, direct voxel manipulation and focus plus context interaction.
In particular, volume visualization can be done with a geometric rendering system which transforms the data into a set of polygons representing an iso-surface. The contour tree algorithm and other alternatives such as branch decomposition are usually used to find these iso-surfaces. Contour tree algorithms may be vulnerable to noise, which can be problematic in luggage inspections since dense materials such as steel cause noise by reflecting the X-rays.
In order to investigate volumetric dataset, one can use the Transfer Function (TF). In practice, this maps the voxel density with a specific colour (including its transparency). Transfer functions can be 1, 2 or n dimensional and are of great help to isolate structures of interest in volumetric data. Thanks to the colour blending process, a suitable transfer function can also reveal iso-surfaces or hide density to improve the volumetric data visualization.
A specific difficulty arises in an environment such as that described with respect to FIG. 1 is that the user's view of a particular article of interest will often be obscured by a number of other objects of no interest. In order to better view the objects of interest, the user may wish to isolate or highlight objects of interest, or selectively remove or render less visible interfering objects. In a voxel based environment however, there is no inherent detection of objects, since the whole environment is simply a matrix of voxels of varying densities. It is desirable to provide functions for effectively identifying discrete objects in such an environment.